Fraud red-flag analytics on expense claims with Benford’s law

Kopano Logistics runs trucks, depots and warehouses across Gauteng, KwaZulu-Natal and the Western Cape. About 640 employees can claim expenses: fuel, tolls, meals, accommodation and small repairs. Claims total about R48 million a year across roughly 30,000 lines. Approval limits are R5,000 for line managers and R20,000 for regional managers.

Last year internal audit found one driver supervisor who had claimed the same fuel slip three times, and a depot manager whose claims were almost all just under his approval limit. Both were found by chance. The audit committee now wants a systematic test that runs every quarter.

The head of internal audit, Mr Sipho Dlamini, has asked you to build it. Because real claims contain personal information, you will generate a synthetic claims file in task 1 with a fixed random seed. A classmate or your mentor should plant some irregular patterns in a copy without telling you where, so you test your method honestly. If that is not possible, plant them yourself before you design any tests and do not look at the key until task 5.

Your work: - first-digit and first-two-digit Benford tests, overall and by claim type; - rule-based tests: duplicates, round amounts, just-below-limit claims, weekend and public holiday claims, split claims and claims after resignation; - a risk score per employee that combines the tests; - a sample of high-risk claims for follow-up with the evidence to request; - a report to the audit committee that is factual, fair and handles personal information carefully.

Done looks like a notebook or workbook that reruns each quarter, a ranked follow-up list and a four-page report. Red flags are reasons to look, not proof of fraud, and your report must say so.